Retros

yat:retro is how the tutor improves. You talk, it asks, and what you notice becomes proposals that a person reviews before anything changes.

What you say

You say	What happens
retro / retrospective	A debrief on the env you're in
run a retro on <env>	A debrief on that env
let's debrief	The same

Four lanes

The debrief walks four questions, one at a time, each grounded in something specific from your files:

1. What clicked, and what made it land?
2. What was hard, and what kind of stuck was it?
3. What strategy emerged: a move, mnemonic or reframe worth naming?
4. What should change about the env itself?

It never answers for you. No input from you means no retro file.

What comes out

• A retro file in memory/retros/.
• Each proposal queued in memory/retros/improvement-queue.md.
• Learnings and mnemonics in the graph.
• Up to three proposed profile lines, each shown with its evidence for you to accept, edit or reject.

Retros never edit skills, env contracts or plans. The retro-reflector agent drafts proposals into memory/retros/proposals/, and a person merges them in git.

Files it touches

Reads	Writes
envs/<env>/progress.md, session-state.md	memory/retros/
envs/<env>/interactions.jsonl, memory/retros/	memory/ontology/graph.jsonl

Terminal: plumbing.

Source

harness/skills/retro/SKILL.md